%% Based on a TeXnicCenter-Template by Gyorgy SZEIDL.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%------------------------------------------------------------
%
\documentclass[a4paper, 12pt, onecolumn]{article}
%%Options -- Point size:  10pt (default), 11pt, 12pt
%        -- Paper size:  letterpaper (default), a4paper, a5paper, b5paper
%                        legalpaper, executivepaper
%        -- Orientation  (portrait is the default)
%                        landscape
%        -- Print size:  oneside (default), twoside
%        -- Quality      final(default), draft
%        -- Title page   notitlepage, titlepage(default)
%        -- Columns      onecolumn(default), twocolumn
%        -- Equation numbering (equation numbers on the right is the default)
%                        leqno
%        -- Displayed equations (centered is the default)
%                        fleqn (equations start at the same distance from the right side)
%        -- Open bibliography style (closed is the default)
%                        openbib
% For instance the command
%           \documentclass[a4paper,12pt,leqno]{article}
% ensures that the paper size is a4, the fonts are typeset at the size 12p
% and the equation numbers are on the left side

\usepackage[left=25mm, right=25mm, top=25mm, bottom=25mm]{geometry}

\usepackage{pdflscape}
\usepackage{amsmath}
\usepackage{afterpage}
\usepackage{epigraph}
\setlength\epigraphwidth{.8\textwidth}
\allowdisplaybreaks[4]

\usepackage[utf8]{inputenc}

\usepackage[maxbibnames=5,maxcitenames=2,citestyle=authoryear,uniquename=mininit,bibstyle=authoryear,natbib=true,autocite=inline,sortcites=true,block=space,isbn=false,url=false,doi=false, backend = biber]{biblatex}

% Define commands/macros for paths that are used so we can switch quickly from the folder structure we work withto the flat folder structure required for submission.
\newcommand{\plotsdir}{../CovidIdeologyReplication/generated_images}
\newcommand{\texdir}{../CovidIdeologyReplication/generated_tex}

\bibliography{bibliography.bib}
\input{psrm_styles.tex}

\usepackage{titling}
\usepackage{verbatim}
\usepackage{comment}
\usepackage{dcolumn}
%\usepackage{epstopdf}
\usepackage{setspace}\setdisplayskipstretch{.5}
\doublespacing
%\onehalfspacing

% https://tug.org/pipermail/texhax/2008-July/010780.html and https://latex.org/forum/viewtopic.php?t=32785
\newcommand{\note}[1]{\footnote{\begin{doublespace} \normalsize #1\end{doublespace}\vspace{-12pt}}}

% https://latex.org/forum/viewtopic.php?t=26814
\usepackage{titlesec}
\titlespacing{\section}{0pc}{6px}{4px}
\titlespacing{\subsection}{0pc}{6px}{4px}

\usepackage[pdftex]{graphicx}
%\usepackage{subfig}
\newtheorem{hypothesis}{Hypothesis}
\newtheorem{treatment}{Treatment}
\usepackage{tikz}
%\usepackage{color,soul}
\allowdisplaybreaks[4]
\usepackage{longtable}
\usepackage{placeins}
\usepackage{booktabs}
\usepackage{amssymb}
\usepackage{subcaption}


\usepackage{array}
\newcolumntype{L}[1]{>{\raggedright\let\newline\\\arraybackslash\hspace{0pt}}m{#1}}
\newcolumntype{C}[1]{>{\centering\let\newline\\\arraybackslash\hspace{0pt}}m{#1}}
\newcolumntype{R}[1]{>{\raggedleft\let\newline\\\arraybackslash\hspace{0pt}}m{#1}}

% From https://tex.stackexchange.com/questions/18017/space-inserted-after-input-call
\newcommand{\inputy}[1]{\input{#1}\unskip}


\usepackage[T1]{fontenc}

\usepackage[%
    colorlinks=true,linkcolor=black,citecolor=black,urlcolor=black,%
    pdfstartview=FitH,%
    bookmarksopen=true,bookmarksopenlevel=5,%
    plainpages=false,pdfpagelabels,%
    pagebackref=false]{hyperref}
\usepackage{cleveref}

\usepackage{xr}
\externaldocument{CovidIdeology_suppl}

% http://tex.stackexchange.com/questions/165128/how-to-stop-texcount-from-counting-begincomment-endcomment
%TC:envir comment [] ignore


\begin{document}

\begin{titlingpage}
%\setlength{\droptitle}{50pt} % lower the title



\newcommand{\thetitle}{The Non-Consequences of COVID-19 on Left-Right Ideological Beliefs}

\title{\thetitle{}}
\author{Jack Blumenau\\
	%School of Public Policy,\\
    University College London\\
	\texttt{j.blumenau@ucl.ac.uk}\\
	%\url{http://www.jackblumenau.com/}\\
	\and
	Timothy Hicks\\
	%School of Public Policy,\\
    University College London\\
	\texttt{t.hicks@ucl.ac.uk}\\
	%\url{http://tim.hicks.me.uk/}\\
	\and
	Alan M. Jacobs\\
	%Department of Political Science,\\
    University of British Columbia\\
	\texttt{alan.jacobs@ubc.ca}\\
	%\url{https://politics.ubc.ca/profile/alan-jacobs/}\\
	\and
	J. Scott Matthews\\
	%Department of Political Science,\\
    Memorial University of Newfoundland\\
	\texttt{scott.matthews@mun.ca}\\
	%\url{https://sites.google.com/view/jsmatthews/home}\\
	\and
	Tom O'Grady\\
	%School of Public Policy,\\
    University College London\\
	\texttt{t.o'grady@ucl.ac.uk}\\
	%\url{https://tomogradypolitics.wordpress.com/}
}

\date{Produced: \today\thanks{Blumenau, Hicks, and O'Grady acknowledge the generous support of their department. Jacobs and Matthews acknowledge the generous support of the Social Sciences and Humanities Research Council of Canada (grant \#435-2014-0603). The April wave of the PACER study and the survey experiment were deemed exempt by the UCL Political Science Ethics Review Committee. Ethics approval for the September wave of the PACER study was granted by the Interdisciplinary Committee on Ethics in Human Research at Memorial University (file \#20210161) and by UBC's Behavioral Research Ethics Board (file H20-01750). Supplementary material for this article is available in the appendix in the online edition. Replication files are available in the JOP Data Archive on Dataverse (\url{https://dataverse.harvard.edu/dataverse/jop}). The empirical analysis has been successfully replicated by the JOP replication analyst.}}
\normalsize

\begin{onehalfspace}
	\maketitle
\end{onehalfspace}


%\end{titlingpage}

%\begin{center}
%\begin{bfseries}
%\thetitle{}
%\end{bfseries}
%$ $\\
%%Word count: XXX
%\end{center}

\begin{singlespacing}
\begin{abstract}
\noindent Scholarship on past major crises, such as wars and depressions, argues that these events transformed mass attitudes about the role of the state. Motivated by these claims, we theorize reasons why the COVID-19 pandemic might have shifted citizens' ideological beliefs and investigate whether it has. Using original panel data from the UK, we find no evidence that the pandemic affected beliefs about the role of government, even for those directly experiencing economic losses or new forms of state relief. In a follow-up survey experiment, we also find that voters do not change their opinions on redistribution or the role of government even when exposed to elite cues that frame the crisis as revealing the need for state expansion. Our findings suggest that crises may more commonly exert their effects on mass beliefs via the long-term feedback effects of elite-driven policy changes than through direct exposure to crisis conditions.% 150 words
\end{abstract}
\end{singlespacing}

\end{titlingpage}

\clearpage

Core political beliefs -- orientations toward basic societal goals and the role of government in society -- tend to be highly stable at the individual level \autocite[e.g.][]{SearsFunk1999, EvansNeundorf2020}. In defense of such attitudes, citizens are generally ``prone to deny, denigrate and actively counter-argue'' \citep[51]{CotterEtAl2020} counter-attitudinal evidence \citep{TaberLodge2006}. Accordingly, attitude change regarding redistribution and the economic role of governments is typically small and short-lived, even following sizable economic shocks \citep{Margalit2013, Margalit2019, Marten2019, NaumannEtAl2016, OGrady2019a, OwensPedulla2014}. Yet past studies of change in left-right ideological beliefs draw on evidence from ``ordinary times'' in political terms, characterized by any or all of: highly stratified exposure to policy information; high levels of elite conflict over salient political issues; little fundamental change in the role of the state; and limited potential for subjectively important inconsistency between the interpretation of new policy information and pre-existing political beliefs. Under such circumstances, a combination of disengagement and resistance is often enough to insulate most voters from internalizing considerations that might permanently change their outlook \citep{Zaller1992}.

On the other hand, historical accounts of extraordinary events, particularly the two World Wars and the Great Depression, argue that they generated broad, enduring mass-level ideological change. Scholars have contended that mass warfare led to leftward shifts in public attitudes toward progressive taxation \autocite[Chap. 6]{ScheveStasavage2015}, that war and depression ``created new understandings of and new attitudes toward governmental action’’ (p.59) and broad support for an expanded state \citep{Higgs2013}, and that these upheavals generated new public expectations of government as a guarantor of basic living standards \autocites{MorganEvans1993}[see also][]{ Titmuss2018}. Analysts have also held that, by producing uncertainty and changes to the distribution of risk, severe crises generate public demand for expanded social insurance \autocite[Chap.8]{Rehm2016} and greater risk-sharing \citep{DryzekGoodin1986}. In some accounts, elite debate and discourse interact with the mass public’s experience of crisis to cause ideological change: \citet{Seabrooke2007}, for instance, argues that World War I and interwar economic shocks reshaped mass expectations in ways that motivated and legitimated elite-generated Keynesian ideas. 

Thus, at one end of the spectrum, we have good evidence that ideological change is rare or fleeting in relatively ordinary times, even amidst sharp economic downturns. At the other extreme, historical studies suggest that the kind of human, physical, and economic devastation wrought by the World Wars and the Great Depression -- together with the massive expansions of state activity that attended these crises -- can bring about broad change in citizens' ideological beliefs, creating political impetus for institutional transformation. Between these two poles, however, lies a great deal of empirical terrain. 

In this paper, we systematically examine the attitudinal impact of a crisis that was far more disruptive and involved a considerably more dramatic expansion of the state  than a typical economic downturn: the COVID-19 pandemic. While the pandemic was less catastrophic than world wars or the great depression, the economic dislocations and human costs that the pandemic generated, especially over its first year, went far beyond those of even a deep recession. Equally important, the pandemic provoked a sudden enlargement of government activity not previously seen outside of wartime.  Never in times of peace have governments asked citizens to sacrifice as much for the common good as they did during the coronavirus outbreak, requiring entire populations to stay at home and putting large segments of the labor force out of work. In turn, most Western governments stepped into the breach -- to a degree not witnessed during the Great Recession or other postwar downturns -- to protect citizens from severe economic losses, including through new or greatly expanded forms of income support, furlough and wage-subsidy schemes, compensation for the self-employed and small businesses, and eviction moratoriums.

Critically, the pandemic also featured informational conditions that, under prevailing political-psychological theories, ought to have been particularly conducive to belief change. The pandemic's direct impacts and associated policy responses were felt extremely broadly and deeply by the general public, with large portions of the citizenry exposed to new information about the distribution of risk and about the capacity of the state to protect them from health and economic perils. And crucially -- at least in the UK, where our paper focuses -- there was little elite conflict about the state's novel interventions during the most acute period of the pandemic. Citizens thus faced an uncommonly one-sided informational environment, one devoid of the discursive resources that might typically have helped more conservative motivated-reasoners defend their existing beliefs.

Is a massive social disruption short of total war -- one that exposes most of the population to new economic and health perils and demonstrates the state's unique capacity to protect against losses -- sufficient to overcome citizens' attitudinal defense mechanisms and shift ideological beliefs to the left? We deploy micro-level observational and experimental evidence to address this question. Focusing on the case of the UK, we ask whether the broad and deep economic rupture generated by the pandemic, and the unprecedented governmental response to it, produced a detectable change in Britons' ideological orientations toward the state. While the UK government's response in the early phases of the pandemic was similar to its OECD peers \citep{EnglerEtAl2021}, the nature and scale of its response to COVID-19's economic effects was decidedly out of keeping with the country's approach to past economic crises and, more generally, the UK's historically liberal pattern of social provision \citep{HanckeEtAl2022}. Moreover, the lack of elite conflict over such unprecedented measures as the government's furlough scheme -- which subsidized the incomes of up to a third of UK workers -- was remarkable. The UK during the pandemic thus presents an ideal setting in which to test conjectures regarding the mass attitudinal consequences of a large, novel, and largely uncontested expansion of state activity.

From April 2020 to May 2021, we fielded survey questions concerning left--right ideological beliefs together with the British Election Study (BES) team in four waves. Because our study uses existing BES panelists, we are able to observe attitudes as far back as 2014, a much longer span of time than other research on the pandemic's attitudinal effects \citep{ReeskensEtAl2021,AresEtAl2021,JuradoKuo2022}.  
We examine individual-level attitude change in general, as well as among those with specific exposure to the crisis. We also consider the possibility that change in mass attitudes arises not merely from direct experience of crisis but from the interpretation of that crisis by elites. Public opinion scholars have long emphasized the role of elite rhetoric as a determinant of political attitudes, particularly via the provision of frames that shape citizens’ understandings of political controversies \citep{ChongDruckman2007}. As \citet{Blyth2002} argues, the meaning of a crisis is often constructed through elite-led processes. We thus fielded a survey experiment to a separate sample that allows us to examine whether citizens respond to discourse that connects key elements of the COVID-19 crisis, including both general features and respondents’ personal experiences, to broader questions about the proper role of the state in society. Seeking to create conditions favorable to an effect, we test four distinct frames linking the pandemic's impact to reasons for an expanded state role while also manipulating the salience of respondents’ exposure to the pandemic’s effects. 

The results, in short, are almost uniformly null. While we find that individuals did update their beliefs about the risk of unemployment and poverty during the pandemic, we find no change, on average, in attitudes toward redistribution, levels of taxation and spending, or government’s responsibility to ensure employment. Leveraging variation in individual-level exposure to various consequences of the pandemic, including government program enrollment, we also compare attitudinal shifts in respondents who were directly affected by the crisis to those among individuals who were not so directly affected, and find that direct exposure made no difference. While we do observe some attitude changes, these are small, inconsistent and isolated to particular subgroups. In the survey experiment, moreover, we fail to find any significant attitude change, even in the presence of a treatment heightening the salience of personal experiences of the pandemic.


Our findings shed light on the limited capacity of even major societal and policy upheavals to change citizens’ understandings of the appropriate role of the state. Moreover, unlike other recent studies of the effect of crisis on ideological beliefs \autocite[e.g.][]{Margalit2013, ReeskensEtAl2021,AresEtAl2021}, we directly examine the effects both of the \emph{losses} that citizens experienced and of the experience of new or expanded forms of state \emph{protection} that the crisis provoked. Importantly, this social-protection mechanism is not merely a separate pathway through which crisis might shape mass beliefs. It also represents a possible \emph{confound} in the search for crises' effects insofar as individuals who benefit from social protection are likely to experience \emph{less} income or job loss than those who do not. Thus, focusing strictly on the effect of economic outcomes may understate the effect of crisis on attitudes, and our original dataset -- with detailed individual-level measures of receipt of social protection -- allows us to distinguish these potential pathways.

Our results overall suggest that it takes more than the ravages of -- and governmental response to -- a global pandemic to move attitudes perceptibly to the left. It is, of course, possible that there are specific features of the COVID-19 crisis that weakened its effects on mass attitudes. Attitudes might well have undergone substantial change if the pandemic's severest impacts and the state's expanded role had endured for much longer. It may also be that the exceptional nature of the shock made it easier for citizens to dismiss its relevance for `normal' politics. Finally, as we explore further in the paper’s conclusion, the absence of an attitudinal shift in the depths of a global pandemic suggests that crisis itself rarely operates as a direct cause of ideological change. Rather than new citizen expectations and demands arising from the experience of upheaval itself, public understandings of the state's role may be downstream of elite-level learning from crisis, changing only as a long-run feedback effect of institutional change.


\section*{Theory}
\label{sec:Theory}

In this section, we spell out a set of theoretical considerations that shape our expectations about whether the pandemic-induced economic crisis and government response might have changed left-right ideological beliefs. We begin by reviewing a set of reasons why, given prevailing models of attitude-formation, new information about economic risks or state capacities might not change citizens' beliefs about the role of government. Second, we consider a set of specific features of the COVID-19 crisis that might have been particularly likely to generate attitudinal change despite cognitive mechanisms that typically make left-right ideological beliefs hard to change. Finally, we outline the content of the information that the crisis and state rescue measures would have conveyed to attentive audiences, including information about economic risk, deservingness of benefit recipients, and capacities of the state to solve social problems.


\subsection*{The Sources of Attitudinal Stability}
\label{sec:Theory:Stability}

Citizens' orientations toward abstract values like freedom and equality and the role of government in society, including what we term left-right ideological beliefs, tend to be strong attitudes that, if not quite immune to change, are highly persistent \citep{SearsFunk1999,EvansNeundorf2020,NeundorfSoroka2018}. As with other strong attitudes, such as party identification, individuals are generally motivated to protect such political beliefs through biased responses to new information. Whereas pro-attitudinal information is often treated uncritically and readily integrated into long-term memory, counter-attitudinal information is subject to considerable scrutiny, often motivating the perceiver to generate counter-arguments \citep{Kunda1990,TaberLodge2006}.

Importantly for the analysis to come, the perceiver's ability to challenge counter-attitudinal information depends on the ease with which pro-attitudinal information can be called to mind. The most knowledgeable partisans, for instance, tend to mount the most successful defense of attitudes with clear partisan implications \citep{Matthews2013}.
 More generally, \citet{TaberLodge2006} find that key mechanisms of motivated political reasoning are conditioned by political knowledge, including the impact of prior attitudes on evaluations of arguments, the denigration of counter-attitudinal information, and pro-attitudinal information-seeking (pp. 760--764). These findings imply an important role for the informational environment in shaping the motivated reasoner's capacity to engage in attitude defense. Simply put, the more the informational environment supplies the ``ammunition with which to counterargue incongruent facts'' \citep[757]{TaberLodge2006}, the easier it becomes to defend one's strong political attitudes from counter-attitudinal information \citep[cf.,][]{JeritBarabas2012}. At the aggregate level, this implies that the informational environment most conducive to stable attitude polarization is one that supplies significant reinforcement for contending perspectives. 

Some citizens, of course, lack strong attitudes even in relation to fundamental political and social principles. Conditional on their receiving a counter-attitudinal political message or other political communication, such individuals are likely to experience attitude change. However, as \citet{Zaller1992} famously argued, precisely that which renders those with weak political attitudes susceptible to political persuasion -- their modest level of engagement with politics -- also generally means they will not receive political communications, owing to lack of exposure, lack of comprehension, or some combination of both. Thus, those citizens whose attitudes are \emph{in principle} most changeable are \emph{in fact} rarely exposed to conditions that would bring about change.

In summary, the stability of political attitudes at the aggregate level, even in the face of substantial social change and political events, is generally thought to be driven by processes of motivated reasoning among the most informed, underwritten by polarized information environments and limited exposure to new information among those with the most labile attitudes. Attitudinal change in the face of economic shocks -- including recessions and financial crises -- thus tends to be limited and short-lived \citep{Margalit2019}.


\subsection*{Distinctive Informational Features of the COVID-19 Crisis}
\label{sec:Theory:DistinctiveInformation}

The distinctive circumstances of the pandemic, however, disrupted the ``normal'' flow of political information and, as such, undermined key conditions typically underpinning attitudinal stability. First, the economic shock and the expansion of government's role were on a far greater scale than in typical recessions and were unlikely to be missed even by the chronically inattentive. The economic crash induced by the pandemic and associated lockdowns was massive, saturating the environment with information about large-scale actual and potential economic losses. The magnitude and structure of the downturn ensured that it had an unusually broad reach, threatening social groups that might have been relatively sheltered in a more ``typical'' recession. Indeed, many of those suffering economic losses from the pandemic will have been surprised, in comparison to past crises, at finding themselves unemployed or at risk of unemployment.\note{See Appendix~\ref{apdx:SubjectiveUnemploymentRisk} for detailed evidence and discussion on this point.} 

Similarly widespread was information about the role of government intervention in mitigating the damage. In comparison to past crises, the pandemic and the state's response drew an unusually large number of people into direct financial dependence on the state. Many individuals who would not have become dependent on the state during a normal recession found themselves suddenly in receipt of payments from the government. In the UK, not only did social benefit rolls swell with the newly unemployed, but a vast program of wage subsidies for furloughed workers paid the wages of millions of people whose reliance on state aid had previously been rather minimal. For salaried employees the furlough scheme covered up to 80\% of their pre-pandemic wage; for the self-employed there was a self-employment scheme of similar generosity. The furlough scheme covered 11.7 million employees and cost \pounds70 billion, with a peak of 30\% of the UK workforce supported, while a scheme to protect the self-employed from income loss cost a total of \pounds28 billion and paid grants to another 2.9 million individuals. Given these programs' extraordinarily broad reach, information about the role of the state in protecting households from financial catastrophe will likely have reached individuals irrespective of prior levels of political interest and knowledge \citep[cf.,][]{LermanMcCabe2017}.\note{See Appendix~\ref{apdx:SubjectiveUnemploymentRisk} for sources on participation in these schemes.} 


Moreover, for the first several months of the pandemic -- the period on which we focus empirically -- elite cues would have created favorable circumstances for the crisis and state response to have generated leftward attitudinal change. In more ordinary times citizens would likely have been exposed to elite arguments against enhanced social protection. These arguments would have helped more conservative motivated-reasoners defend their existing beliefs. However, such discursive supports were absent in the first phase of the crisis, which witnessed broad elite consensus on the necessity of expanding the social safety net. In fact, the elites who might normally have critiqued a massive expansion of social protection, Conservative politicians, were the authors of this policy change. Most Conservative politicians surely hoped that social spending would return to its pre-pandemic levels as soon as politically feasible. But, importantly, they weren't \emph{talking} about it -- making anti-state-expansion arguments -- in public during this time. The informational environment up to mid-2020 was essentially bereft of prominent elite messages that would have facilitated motivated reasoning in defense of small-government attitudes.



\subsection*{The Content of Crisis-Induced Information}
\label{sec:Theory:ContentInformation}

Having argued that the pandemic generated unusually large (and uncontested) flows of political information, we now discuss the content of that information, focusing on three features that might have had implications for citizens' left-right ideological beliefs.

\textit{1. Information about economic risk.} The broad economic losses arising from the pandemic plausibly conveyed new information about the distribution of economic risk, with the pandemic's adverse material impacts rising much further up the income scale than those experienced during a normal recession.

\citet{RehmEtAl2012} demonstrate that the correlation of labour market risk and income is strongly negative for the UK, meaning that higher-income workers usually face the lowest risks of unemployment. Yet the COVID-induced shock was so powerful that its effects spread relatively high up the income distribution, affecting people who are usually quite insulated from downturns. We can think about why this might have mattered in light of the well-established ``availability bias'' in human cognition, which leads individuals to rate as more likely those outcomes that they can more readily imagine or call to mind \autocite{TverskyKahneman1973}. News or direct experience of dramatic, widespread economic losses during the COVID-induced recession should have made adverse economic outcomes more cognitively available to citizens who are typically sheltered from job and income loss, leading to upward updating of subjective economic risk. 

The economic devastation wrought by COVID-19 could, therefore, have generated a shift in preferences through change in perceived individual objective interests. As a number of prior studies have demonstrated \citep{ChongEtAl2001, Haselswerdt2020, OGrady2019a}, self-interest tends to play a greater role in shaping policy attitudes, compared to symbolic considerations, when the material stakes are large, clear, and salient.  Even where individuals did not directly experience loss or benefit-recipiency themselves, a sense of social affinity with those who did should have also mattered: \citet{StapelEtAl1994} find that individuals judge as more likely those events that they observe to have happened to other individuals who belong to the same social category as they do. By generating large job losses across a wide range of income and occupational groups, the COVID-induced recession may have made the general risk of labor-market losses imaginable and personally relevant to an especially large number of individuals. Higher market-risk assessments should, in turn, generate greater demand for social protection \autocite{Rehm2016}.

\textit{2. Demonstration of unique state capacities.} Prevailing narratives about the state in recent decades have primarily focused on its decline. Globalization and financialization have often been depicted as implying a diminished capacity of the state to shape economies and societies. Even the most recent, similarly-sized economic shock -- the financial crisis and ``Great Recession'' of 2008--9 -- appeared to reveal governments as being at the mercy of bond markets and central-bank technocrats \autocite[e.g.][]{BrazysHardiman2015}. 

The pandemic, and the widely visible policy responses that it demanded, had the potential to reverse this view. In both public health and economic terms, the pandemic plausibly provided a powerful demonstration of the capacity of the state to act as the ultimate guarantor of security for its population. The economic response involved much more than just an expansion of existing social programs. It involved the rapid adoption of \textit{new} forms of state intervention. The transition from ``there is a new virus in China''  to ``make sure you wash your hands'' to ``almost everyone must stay at home and the state will backstop your income'' was as quick as it was unimaginable only weeks prior to the initial outbreaks. Moreover, no actor other than the state had the power to respond comprehensively to the problem. To the extent that anti-statist attitudes derive, in part, from beliefs about the diminished capacities of the state, or from the notion that the state has been functionally supplanted by market or non-governmental actors, the pandemic response served as a potentially dramatic demonstration of the state's strength and capacities to provide relief and generate collective action.

While this process may have affected all citizens, those receiving government aid for the first time may have been especially affected. Work in the US has shown not only that enrollment in a program affects attitudes toward that scheme and others in the same policy domain \citep[e.g.,][]{LermanMcCabe2017, HopkinsParish2019}, but also that citizens' experiences with individual social programs can generalize, shaping overall perceptions of the state \citep{Mettler2011, Soss1999}. The direct receipt of state benefits may have dramatized for some UK citizens, for the first time, the value of the state as an insurance mechanism against shocks and a source of relief for those falling on hard times. 


\textit{3. Deservingness information.} A large body of evidence points to the role of beliefs about deservingness -- especially, the degree to which recipients' need arises from circumstances beyond their control -- in shaping support for social welfare spending \autocite[e.g.][]{SnidermanTetlock1991, Gilens1999, Petersen2012, Fong2001}.
 Yet, those negatively affected by the COVID-induced labor-market shock were very obviously not to blame for their unfortunate circumstances. This means that, in comparison to other circumstances generating financial need or unemployment, the victims of the COVID-induced recession may have been looked on with an unusual degree of sympathy. Consistent with this, in his study of representations of poverty in the U.S., \citet[127]{Gilens1999} finds that, during times of general economic distress, coverage of poverty is both more sympathetic and more likely to connect the circumstances of the poor to national conditions. It is possible that beliefs about deservingness formed in the context of the COVID-19 crisis \autocite[e.g.][]{BridgmanEtAl2022} might generalize, yielding a shift toward more favorable beliefs about the deservingness of welfare recipients and greater support for welfare in general. 

The broad dispersal of public largesse may also have affected deservingness perceptions. Following the logic of our motivated-reasoning argument, conservative citizens who became new beneficiaries of state support were likely to want to reconcile that new status with their desire for a positive self-concept. The dissonance between accepting state support and seeing the welfare state as wasteful and colonized by the undeserving ought, in principle, to have been fairly striking -- with a potential resolution of this dissonance via a softening of views on welfare and state support more broadly. Moreover, there is considerable evidence that attitudes toward redistribution are shaped by individuals' sense of social affinity with beneficiaries \autocite{CavailleTrump2015, LupuPontusson2011}. To the extent that recipients during the pandemic represented a more socially diverse group, relative to the ``typical'' group of benefit recipients, a broader range of citizens -- even those who did not themselves receive benefits -- may have felt a sense of social proximity to beneficiaries and, thus, perceived them as deserving. These shifts in deservingness perceptions, in turn, may have increased support for social spending. 


\section*{Survey Data}
\label{sec:Data}

We investigate the impact of the pandemic on UK citizens' ideological beliefs using data from the British Election Study (BES) Internet Panel (2014--2021) and our own Public Assessments of COVID-19 Economic Response (PACER) survey. Waves 1--19 of the BES provide pre-crisis observations for more than 30,000 respondents. We observe some of these respondents on three further occasions during the COVID-19 pandemic. The first and third of these survey waves come from the PACER surveys, fielded by YouGov in April 2020 ($n=3041$) and September 2020 ($n=3149$) to subsets of BES panelists. The second crisis survey wave comes from the BES wave 20, fielded in June 2020 ($n=31468$).

All respondents appearing in the PACER waves, and most of those in BES wave~20, were drawn from wave~19 of the BES, allowing us to track them from the pre-crisis period through the first phase of the pandemic. Altogether our data include information on 32,352 individuals from the BES panel who appear in at least one survey wave after the start of the crisis. 28,997 respondents appear in just a single crisis wave, 1,404 appear in just two crisis waves, and 1,951 respondents appear in all three crisis waves.  In addition, although the questions that form the basis of our key subgroup treatment variables were not included in subsequent BES waves, we also present aggregate data below from wave~21 of the BES, fielded in May 2021 ($n=30281$). Our aggregate data therefore span four coronavirus waves and stretch over a year into the pandemic.

We assess impacts on three outcome variables as measures of left-right ideological beliefs. First, $redistSelf_{i,t}$ focuses on economic inequality: ``Some people feel that government should make much greater efforts to make people's incomes more equal. Other people feel that government should be much less concerned about how equal people's incomes are. Where would you place yourself on this 0-10 scale?''. Second, $taxSpendSelf_{i,t}$ addresses the size of government in fiscal terms: ``Using the 0 to 10 scale below, where the end marked 0 means that government should cut taxes a lot and spend much less on health and social services, and the end marked 10 means that government should raise taxes a lot and spend much more on health and social services, where would you place yourself?''. Lastly, $jobsForAll_{i,t}$ focuses on the economic role of government: ``People have different views about society and the economy. How much do you agree or disagree with the following statements?\dots `It is the government's responsibility to provide a job for everyone who wants one'.'' In all analyses, DVs are scaled such that higher values correspond to more left-wing responses. Further, we standardize the DVs to have mean of 0 and standard deviation of 1 to ease effect comparisons.

All three DVs were measured in all pandemic waves (PACER~1, BES~20, PACER~2, BES~21); $redistSelf$ and $taxSpendSelf$ were also asked in waves preceding it, back to 2014 and 2018, respectively. Thus, our estimates of the effect of exposure to crisis-induced shocks on $jobsForAll$ are based only on variation within the crisis, whereas estimates of effects on $redistSelf$ and $taxSpendSelf$ also leverage pre-crisis observations.\note{See figure~\ref{fig:dv_by_wave} for details of survey-wave coverage for each dependent variable and section~\ref{apdx:PanelQWordings} for full question wordings.} 


\section*{Aggregate Change in Left-Right Ideological Beliefs}
\label{sec:MacroTemporal}

%\afterpage{
%\begin{landscape}
\begin{figure}[htp]
	\centering
	\includegraphics[width=.48\textwidth]{\plotsdir/fig1_redistSelf_by_wave.pdf}
	\includegraphics[width=.48\textwidth]{\plotsdir/fig1_taxSpendSelf_by_wave.pdf}
	%\includegraphics[width=.9\textwidth]{\plotsdir/ideoBHPSJjobForAll_by_wave.pdf}	
	\caption{Average value of each outcome variable over time (higher = more left-wing response). Averages are weighted using cross-sectional BES/YouGov survey weights.}
	\label{fig:ideo_var_by_wave}
\end{figure}
%\end{landscape}
%}

Did the left-right ideological beliefs of UK citizens change in aggregate during the COVID-induced economic crisis? Figure~\ref{fig:ideo_var_by_wave} shows the (weighted) average value of $redistSelf$ and $taxSpendSelf$ over time from the pre-pandemic period to the waves fielded during the pandemic. There is, overall, little evidence of large-scale aggregate change. The left panel in figure~\ref{fig:ideo_var_by_wave} suggests that the UK public became no more favorable towards the idea of redistribution during the crisis. Similarly, the right panel shows no evidence that our respondents became more desirous of higher levels of taxation and public spending. In fact, they became slightly less supportive of it during the pandemic, although overall support was virtually unchanged compared to 2018.\note{In section~\ref{apdx:DiffAggByPrior}, we show evidence of a discernible leftward change on $taxSpendSelf$ for the subset of respondents defined as ``right-wing'' by their pre-pandemic $taxSpendSelf$ attitudes. However, the small size of this group, the lack of a parallel pattern for $redistSelf$, and the conflation of the $taxSpendSelf$ question wording with ``health services'' \autocite[see][]{BlumenauEtAl2022} lead us to doubt that this constitutes noteworthy change in ideological beliefs.} Lack of data availability leaves us unable to make comparisons with the pre-pandemic period for the $jobsForAll$ variable. That said, there was a small increase in support for the government providing jobs during the early part of the pandemic, some of which was later reversed (analysis not reported). 


\section*{Looking for Subgroup Effects}
\label{sec:limitedeffects}

Given the absence of substantial aggregate ideological shifts during the pandemic, we now drill deeper, looking for effects among subgroups whose opinions ought to have been most likely to change: those who personally experienced losses or benefit receipt, those who received more visible forms of assistance, and those from particular partisan groups.

\subsection*{Personal Experience}
\label{sec:limitedeffects:personalexperience}

We have proposed that ideological-belief change may have been especially strong among (i) those who directly experienced pandemic-related economic losses and (ii) those who were in direct receipt of government assistance as a result of the pandemic. To test these possibilities, we employ data from questions put to respondents about a range of personal economic experiences during the pandemic. This included respondents' status in the labor market, use of pre-existing state benefits (including Universal Credit), and receipt of employment or income support from either of the new governmental schemes (the furlough and self-employment support schemes). For all state-benefits questions, we asked about both personal experiences and experiences of other household members. We combine these various responses to capture whether respondents \emph{or} one of their household members experienced an economic change or received a benefit. 

We use responses to these variables to define the following treatment dummy variables, where each is set to 0 for all observations from pre-pandemic survey waves: $CovidFurlough_{i,t}$=1 if the respondent reports that they or a household member became a beneficiary of the ``Coronavirus Job Retention Scheme'' (the furlough scheme for employees); $CovidSelfEmployment_{i,t}$=1 if the respondent reports that they or a household member became a beneficiary of the ``Self-Employed Income Support Scheme''; $CovidUniversalCredit_{i,t} $=1 if the respondent reports that they or a household member received, ``as a result of the coronavirus outbreak,'' Universal Credit or some ``other state benefit'' (not including the furlough or self employment schemes); $CovidGovSupport_{i,t}$=1 if the respondent reports that they or a household member was a beneficiary of \emph{any one of} the preceding three programs; $CovidLostEmployment_{i,t}$=1 if the respondent reports that their working hours had declined relative to before the crisis (including becoming unemployed).

Our goal is to analyze whether respondents with direct experience of economic impacts of the crisis, and the associated state response, were more likely to shift their political views than those without direct experience. The models we estimate rule out a number of sources of omitted variable bias that might otherwise be of concern in this setting. First is the possibility that individuals who were directly affected by the COVID crisis are different from those who were not affected, and that those differences are correlated with political attitudes. Our individual-level fixed-effect approach allows us to rule out omitted variable bias stemming from any such differences that are fixed over time. Second, wave fixed-effects rule out common shocks that might affect the responses of all respondents in a given period. They allow us to capture differences in attitude change between treated and untreated respondents controlling for aggregate change in attitudes.
 
The inclusion of individual fixed-effects means that identification relies on within-individual variation in our treatment variables. Table~\ref{tab:treat_dist_by_wave} shows the share of treated respondents in each of the three crisis waves. These proportions are relatively stable across crisis waves. Notably, little new unemployment was reported, in part because so many people in pandemic-hit sectors were supported by the furlough scheme.

\tiny
\vspace{.5cm}
 \input{\texdir/treatment_variation.tex}
  \normalsize

In this part of the paper we use a baseline specification that estimates the effects of employment loss and receiving any form of government support: 
\begin{eqnarray}\label{eq:fe_model}
	Y_{i,t}	&	=	&	\beta_1 CovidGovSupport_{i,t} + \beta_2 CovidLostEmployment_{i,t}  + \alpha_i + \delta_t + \epsilon_{i,t}
\end{eqnarray}
\noindent where $Y_{i,t}$ is the outcome (i.e. $redistSelf$, $taxSpendSelf$, or $jobsForAll$) for respondent $i$ in wave $t$, and $\alpha_i$ and $\delta_t$ are fixed-effects for respondent and survey wave, respectively.

With all dependent variables standardized and higher values indicating more left-wing preferences, the coefficients of interest -- $\beta_1$ and $\beta_2$ -- represent the degree to which personal exposure to different dimensions of the crisis is associated with leftward shifts in respondent attitudes, measured in standard deviations of the outcome variables. More specifically, the coefficients capture the effects of respondents' receipt of government assistance or lost employment during the period of the crisis, relative to respondents who received no such assistance or did not lose employment.\note{Our fixed-effect design relies on an assumption of parallel trends between treatment and control respondents: that is, that individuals whose financial situation was not affected by the pandemic serve as a reasonable counterfactual for those who were more affected. In section~\ref{apdx:Exp:ParallelTrends}, we demonstrate the plausibility of this assumption in our setting. We show that, for both $redistSelf$ and $taxSpendSelf$ (the two variables yielding multiple waves of pre-crisis data), the trends for those who lost employment and those who did not, and those who received government support and those who did not, are very similar. In addition, in section~\ref{apdx:Exp:Attrition} we discuss the potential for differential panel attrition between treated and non-treated respondents to cause bias in the estimates that we report here.}

The results are shown in figure~\ref{fig:baseline_ideoresults}. These estimates provide little evidence of belief-updating as a consequence of direct economic and policy experiences. At a 10\% significance level, those who received government assistance became more supportive of redistribution ($redistSelf$), but not of higher levels of taxation and spending ($taxSpendSelf$) or of government providing jobs ($jobsForAll$), although the latter coefficient is in the expected direction. Point estimates are also very small -- only a few percentage points of a standard deviation. We also find no clear evidence that losing employment during the pandemic shifted people's attitudes (see Figure~\ref{fig:visibility_ideoresults}), though (as noted) we have low variance on this treatment variable. Nor do we find stronger treatment effects either among respondents with weaker prior attitudinal attachments -- those who provided a ``Don't know'' response in pre-pandemic waves -- as shown in \Cref{fig:subgroups_dk} or among those reportedly receiving government labor market support for the first time (\Cref{fig:subgroups_prior_gov_support}).\note{In section~\ref{apdx:JobLossBenefitInteraction}, we evaluate whether the effects of job loss on attitudes are conditional on whether the respondent was also a recipient of (non-pandemic-specific) state benefits. 
	Receiving state benefits has no moderating effects on $redistSelf$, $jobsForAll$, or $riskPoverty$. While significant interactions emerge for $taxSpendSelf$ and $riskUnemployment$, these are informed by small numbers of observations.}  


\begin{figure}[htp]
	\centering
	%\hspace{-2cm}
	\includegraphics[width=0.6\textwidth]{\plotsdir/fig2_fe_baseline_ideo_vars.pdf}
	\caption{Ideological effects of Crisis Exposure: Estimates of $\hat{\beta}_1$ with 95\% Confidence Intervals. See table~\ref{tab:baseline_ideoresults} for full results.}
	\label{fig:baseline_ideoresults}
\end{figure}

In addition, in section~\ref{apdx:Exp:LongRun}, we use data from wave~21 of the BES to investigate whether the small effects documented here are attributable to the relatively short post-crisis time period we study. Although we cannot use exactly the same measures of crisis exposure for wave~21, results for this later period are strikingly similar to those presented here.


\subsection*{Visibility}
\label{sec:limitedeffects:visibility}

Next, we consider the possibility that the nature of the information generated by receipt of state support may have varied across programs. A rich literature on accountability and welfare-state politics emphasizes the importance of \textit{visibility} in shaping public responses to state action. Citizens are unlikely to credit the state for the benefits they receive from public programs if they cannot easily observe the delivered gains and trace them back to state action \citep{Pierson1993, TilleyEtAl2018}. \citet{Mettler2011} presents micro-level evidence from the United States that social programs that deliver benefits indirectly, such as through private-sector intermediaries, are less likely to be recognized by recipients \textit{as} public benefits and that citizens are less likely to form opinions about ``submerged'' state programs that cohere with their own material interests. More generally, as noted above, work on the impact of ``self-interest'' on policy attitudes suggests such effects are more likely when a policy's personal costs and benefits are highly visible \citep[e.g.][]{ChongEtAl2001}.

The UK government's COVID-relief programs varied markedly in how directly they delivered their benefits. While Universal Credit involves a direct government payment to individuals, and those receiving the self-employment support had to actively register for it, the furlough subsidies were paid to \emph{employers}, who in turn continued to pay their employees as before. The state's hand in supporting employment in the furlough program is, thus, likely to be harder for beneficiaries to detect and less salient than the state's role in Universal Credit or Self-Employment support. In turn, the attitudinal impact of the government's economic response may be larger for those benefiting from one of the latter two schemes than for those benefiting from the furlough program.\note{To be clear, respondents reporting being furloughed knew that they were receiving some kind of assistance. Yet, as Mettler's research shows, awareness of receipt of a benefit does not imply awareness of benefiting from \emph{state} support: U.S. beneficiaries of mediated social programs, even when they knew they had received support, were less likely to view the state as their benefactor than were beneficiaries of direct state provision.}

Figure~\ref{fig:visibility_ideoresults} shows fixed-effects coefficient estimates from a regression that omits $CovidGovSupport$ and instead includes $CovidFurlough$, $CovidSelfEmployment$ and $CovidUniversalCredit$. We find no evidence of a visibility effect. If anything, evidence of a program-specific effect is strongest for the furlough scheme despite this being the least visible program. All other estimates are small and mostly statistically indistinguishable from zero. The visibility of a program does not seem to determine whether it shifts attitudes, and for none of the schemes did benefit receipt have much discernible effect.


\subsection*{Attitudinal Starting Points: Party Support}
\label{sec:limitedeffects:partyid}

Finally, we consider the possibility that citizens' attitudinal starting points -- their pre-pandemic left-right ideological beliefs -- might have limited the moveability of their attitudes. For one thing, supporters of different parties may respond to the same information in different ways -- although the literature does not suggest clear predictions of exactly how \citep{Margalit2019}. \citet{BrooksManza2013}, studying the Great Recession in the U.S., found that Republicans shifted against government intervention during the crisis. \citet{Margalit2013}, on the other hand, found that more right-wing people who lost their jobs during the crisis shifted in favor of more social spending, but left-wingers largely did not. It is also possible that ``ceiling effects'' constrained attitudinal movement. Existing left-wingers might not have been moveable, even in the face of severe economic loss or experience of government relief, simply because they did not have much room to move further left. Consistent with this idea, \citet{Wehl2020}, using German Socio-Economic Panel data, found that the attitudinal effects of employment-status changes depended on respondents' starting points: adverse shocks tended either to reinforce the attitudes of those already on the left or to affect those respondents without strong prior predispositions. 
 
To investigate partisan differences in responses to the pandemic, Figure~\ref{fig:partyid_ideoresults} shows the results from regressions that are the same as those in Figure~\ref{fig:visibility_ideoresults}, but with the inclusion of variables measuring party support, as well as interactions between party support and the receipt of the different types of public benefits. Party support is measured using the respondents' recalled vote in the 2019 General Election, as reported in panel waves after the election. We focus on Labour, Conservative and other party supporters. We are unable to assess the effect of employment loss in this context because the size of partisan groups for those who lost employment is too small for reliable estimation. 

\begin{figure}[htp]
	\centering
	%\hspace{-2cm}
	\includegraphics[width=0.6\textwidth]{\plotsdir/fig3_fe_by_party_by_treat.pdf}
	\caption{Effects of Crisis Exposure on Left-Right Ideological Beliefs for Supporters of Different Parties (coefficients and 95\% Confidence Intervals). See table~\ref{tab:fe_by_party_by_treat} for full results.}
	\label{fig:partyid_ideoresults}
\end{figure}

Figure~\ref{fig:partyid_ideoresults} demonstrates that Labour supporters and supporters of ``other'' parties were very slightly more likely to shift to the left than Conservative party supporters, but statistically significant shifts were rare even in these subgroups. 
Conservative supporters, on the other hand, became very slightly \emph{less} left-wing across some outcome variables and treatments, but for the most part their views did not change either. Thus, when updating occurred in the expected direction, it occurred only for non-Conservative voters, in direct contradiction of the ``ceiling effects'' hypothesis. At the same time, we also note that, even just for Labour supporters, the pattern of effects is quite inconsistent across program types and survey items, providing little overall support for the claim that exposure to pandemic-induced shocks yielded partisanship-conditioned change in ideological beliefs.


\section*{Explaining Non-Effects}
\label{sec:Noneffects}

We turn now to exploring why we observe quite minimal effects of even personal loss and benefit-receipt on left-right ideological beliefs. In our theoretical discussion, we argued that one mechanism through which the crisis might have changed ideological orientations is by changing individuals' perceptions of their risk of future economic loss. We also theorized that crisis might shift left-right attitudes by changing beliefs about the deservingness of benefit recipients. In this section, we look for a possible break in each of these theorized causal chains by examining the effects of the crisis on the relevant mediating beliefs: perceptions of economic risk and views of deservingness.

\subsection*{Changes in Risk Perceptions}
\label{sec:PersonalExperience:risk}

We hypothesized that experiences of economic loss might affect left-right ideological beliefs via changes in individuals' perceptions of their own exposure to the risks from which state programs can provide protection. Did experience of the COVID-induced economic crisis \textit{fail} to change attitudes because it did \textit{not} alter these risk perceptions? To assess this possibility, we use survey items asking respondents about the likelihood of experiencing adverse economic outcomes in the future.  One item ($riskPoverty$ ) asks how likely respondents think it is that during the next year they will ``not have enough money to cover day to day living costs'' while the other ($riskUnemployment$) asks how likely it is that they ``will be out of a job and looking for work.'' 

\begin{figure}[t]
	\centering
	\includegraphics[width=.48\textwidth]{\plotsdir/fig4_riskUnemployment_by_wave.pdf}
	\includegraphics[width=.48\textwidth]{\plotsdir/fig4_riskPoverty_by_wave.pdf}
	\caption{Aggregate change in perceptions of unemployment- and poverty-risk over time (higher=more risk). Averages weighted using cross-sectional BES/YouGov survey weights.}
	\label{fig:risk_var_by_wave}
\end{figure}

There is some evidence that the crisis shifted risk-perceptions in the aggregate, but only temporarily: in Figure~\ref{fig:risk_var_by_wave}, we observe in the April 2020 PACER wave a large aggregate increase in respondents' perceived risk of becoming unemployed over the next year. However, this change was fleeting, with perceived unemployment risk quickly reverting to pre-crisis levels in subsequent waves. Indeed, fear of unemployment fell to its lowest level in the series in the final BES wave in May 2021. The April 2020 wave showed a tiny increase in perceived risk of being unable to cover daily living costs in the next 12 months, but this perceived risk subsequently \emph{fell} over the subsequent three waves (possibly because those who remained employed saved during the lockdown).


\begin{figure}[t]
	\centering
	\hspace{-2cm}
	\includegraphics[width=0.6\textwidth]{\plotsdir/fig5_fe_baseline_risk_vars.pdf}
	\caption{Effects of Crisis Exposure on Risk Perceptions (coefficients and 95\% Confidence Intervals). See table~\ref{tab:baseline_risk_results} for full results.}
	\label{fig:risk_results}
\end{figure}

However, what about those who (or whose households) received government help or suffered loss of employment? Did these personal experiences lead to updating of risk perceptions? Figure~\ref{fig:risk_results} plots coefficient estimates from fixed-effects regressions identical to those in the Personal Experience analyses, except that the dependent variables are now $riskPoverty$ and $riskUnemployment$. Response scales are standardized to have mean 0 and standard deviation of 1. In all cases, we find strong evidence of updating in the expected directions. People who received government support or lost employment \textit{did}, subsequently, consider themselves at greater risk of experiencing poverty or unemployment in the next year. These changes are generally large, up to almost half of a standard deviation. We can thus \textit{rule out} a lack of change in risk perceptions as an explanation for the lack of change in ideological beliefs among those with personal experiences of economic loss or benefit receipt. 
%\newpage


\subsection*{Changes in Deservingness Perceptions}
\label{sec:Deservingness}

We have also argued that the pandemic might have changed attitudes by making individuals see benefit recipients as more deserving, whether through direct experience of loss or benefit receipt or through a feeling of greater social affinity toward those receiving benefits. To investigate this mechanism, we examine responses to two further survey items: $reasonForUnemployment_{i,t}$ queries level of agreement with the statement ``When someone is unemployed, it's usually through no fault of their own,'' and $govtHandouts_{i,t}$ asks agreement with ``Too many people these days like to rely on government handouts.'' Both were rescaled to have a mean of 0 and standard deviation of 1, and the $govHandouts$ variable was reverse-coded such that higher values indicate stronger perceptions of deservingness. Both were asked in some pre-pandemic waves, albeit substantially further in the past than for our other variables. 

The aggregate evolution of these two variables over time (on the original response scale) is displayed in Figure~\ref{fig:deserve_var_by_wave}.  There is some evidence of increased perceptions of deservingness \emph{during} the pandemic, particularly for $reasonForUnemployment$, although it only returned to its 2015 level. Between the last pre-crisis observation in 2016 and the onset of the pandemic, there was no overall change in $reasonForUnemployment$ but a substantial leftward shift is evident for $govtHandouts$. The lack of data in our panel from 2017--19 means we cannot say whether this shift is due to the pandemic itself or occurred earlier. However, in section~\ref{apdx:BSAS} we use comparable data from the British Social Attitudes Survey to show that most of the leftward shift occurred before the pandemic, so that the pandemic itself had, at best, a very minor impact on aggregate deservingness perceptions.

\begin{figure}[ht]
	\centering
	\includegraphics[width=.48\textwidth]{\plotsdir/fig6_reasonForUnemployment_by_wave.pdf}
	\includegraphics[width=.48\textwidth]{\plotsdir/fig6_govtHandouts_by_wave.pdf}
	\caption{Aggregate change in deservingness views over time (higher = more deserving)}
	\label{fig:deserve_var_by_wave}
\end{figure}

\begin{figure}[ht]
	\centering
%	\hspace{-2cm}
	\includegraphics[width=0.6\textwidth]{\plotsdir/fig7_fe_multi_treatment_deserving_vars.pdf}
	\caption{Effects of Crisis Exposure on Deservingness Perceptions (coefficients and 95\% Confidence Intervals). See table~\ref{tab:fe_multi_treatment_deserving_vars} for full results.}
	\label{fig:deserveresults}
\end{figure}

Did those who experienced economic loss or received state help undergo change in deservingness beliefs? Figure~\ref{fig:deserveresults} shows coefficients from the fixed-effects regressions parallel to those in the Visibility section, estimating the effects of each of the three state programs and employment loss on $reasonForUnemployment$ and $govtHandouts$.  Deservingness perceptions were largely unaffected by any of these four treatments. Even when statistically significant, all estimated effects are very close to zero. The only treatment for which we find anything close to substantial updating is loss of employment. 

In sum, in seeking to explain the lack of ideological-belief change, we find mixed evidence of \textit{aggregate} change in two likely mediators of the effect. The effects of personal experience differ, however, between the mediators: while deservingness beliefs were unmoved, respondents who lost work or received benefits \textit{did} perceive greater future economic peril. The lack of ideological movement among those personally impacted by the crisis, in short, does not seem attributable to a lack of updating of risk perceptions. Those exposed to COVID-related economic and policy shocks updated their beliefs about economic risk \textit{without} updating their beliefs about the appropriate role of the state.

\section*{The Role of Political Discourse: Evidence from a Survey Experiment}
\label{sec:SurveyExperiment}

We now probe this puzzle further by examining whether the pandemic might have created conditions \textit{conducive} to attitude change \textit{had} elites supplied \textit{frames} linking experience of the pandemic to more general claims about the appropriate role of the state.  In fact, a notable feature of the pandemic in the UK, especially its early stages, was a lack of strong messaging from any party, including Labour, arguing that the pandemic would or should lead to long-term changes in the role of government or welfare policies. The failure to advance such messaging was, arguably, part of the overall pattern of elite consensus regarding the pandemic economic measures: that is, just as rhetorical opposition to the measures was politically problematic in the harrowing early months of the pandemic, so may have been arguments that leveraged the crisis to advance a broader political agenda regarding the economic role of government. It may, in other words, have been perceived as unseemly to be seen capitalizing politically on a global pandemic.

To test whether elite messaging can induce attitudinal change among a population that has experienced the conditions generated by the pandemic, we fielded a survey experiment that randomly exposed people to different types of discourse that British politicians could plausibly have employed --- to connect the experience of the pandemic to larger ideological principles --- during the period covered by our panel. The experiment was fielded, in August and September, 2021, to a sample of 2,500 British adults recruited through the survey firm Lucid. The sample was balanced against the UK population in terms of the joint distribution of age-group and sex, as well as region and education, and we restricted our sample for analysis to respondents who passed two questions measuring attentiveness to the survey.\note{Our pre-analysis plan contains power calculations justifying the choice of sample size. We used six age bands, the 12 NUTS level 1 regions for the UK, and three education levels (GCSE or equivalent and below, A-Level or equivalent, degree level or higher.)} The survey design, hypotheses, and empirical tests presented in this section follow, without deviation, a pre-analysis plan registered in advance through the Open Science Framework. 

We began without strong priors on what specific \emph{type} of discourse would be most likely to affect opinions, and therefore investigated several different possibilities. We designed four short passages that made different types of arguments, all of which advocated in some way that the crisis should lead to a larger economic role for the government, whether intervening more broadly in the economy or making benefits and national insurance schemes more generous. Table~\ref{tab:SurveyPrimes} contains a brief summary of each of these four primes. The full texts are contained in Section~\ref{apdx:Exp:Primes}. Respondents assigned to read one of these four primes were told ``the coronavirus pandemic has prompted debate about whether the UK government should play a different role in the economy going forward. We will now show you an example of an argument that has been put forward in this debate. Please read this argument carefully.'' They were then asked how strongly they agreed or disagreed with the statements. Respondents assigned to a control condition did not read any arguments.\note{These four arguments are wide-ranging, but do not exhaust \emph{all} possible rhetorical arguments. As suggested by a reviewer, politicians could have argued economic reform was needed to prevent excess profits from online sales during lockdowns. This may have been consequential as \textcite{ScheveStasavage2015} argue that rhetoric about ``profiteering'' during wartime helped politicians subsequently argue for progressive taxation.} 

\begin{table}[htp]
\centering
\small
\begin{tabular}{ll}
\hline
\hline
Prime & Argument \\
\hline
Common Risk and & Covid demonstrated how vulnerable we are to risk, so\\
Insurance Prime (CRIP) & we need stronger national insurance and benefits\\
\hline
Unequal Risk and &  Covid demonstrated how unequal risks are in our society,  \\
Insurance Prime (URIP) & so we need stronger national insurance and benefits\\
\hline
State Capacity & Covid demonstrated the government's ability to solve\\
Prime (SCP) &  problems, so it should play a larger role in society in future\\
\hline
State Incapacity & Covid demonstrated that underfunding our government leads \\
Prime (SIP) & to problems, so we need to invest in state capacity in future \\
\hline
\hline
\end{tabular}
\caption{Summary of the Experimental Primes}
\label{tab:SurveyPrimes}
\end{table}


Finally, we also hypothesized that the strength of the effects of the various types of discourse would depend on the salience of the respondent's own exposure to effects of the crisis. Therefore, half of the respondents were primed to think about the effects of the crisis by being asked -- before reading the argument about the pandemic -- whether they or someone close to them had lost their job, experienced a reduction in income, or had a business fail, and whether they had used the forms of state support that we measured in our panel data. Assignment to this ``personal COVID experience'' prime (PCE), orthogonal to frame-assignment, yields ten different experimental groups.

Following recent methodological work \parencite{CliffordEtAl2021}, we employed a pre--post design in which respondents were asked the dependent variable questions before and after the treatment, with the aim of using the pre-treatment measures as covariates to increase estimation precision. Following \citet{CliffordEtAl2021}, between the two sets of dependent variable questions, we inserted a block of distractor questions that asked about issues unrelated to the experiment, such as mathematical puzzles.\note{The flow of the survey experiment for each group is summarized in Table~\ref{tab:SurveyProtocol}.}

Our pre- and post-treatment batteries of dependent variable questions included the three questions from our panel study: $redistSelf$, $taxSpendSelf$ and $jobsForAll$. To enhance the likelihood of picking up an effect, we pose additional questions relevant to left-right ideological attitudes and use them to construct indices capturing beliefs about redistribution and about the role of government in the economy. Using principal component (PC) analysis, we generate a measure of redistribution beliefs using responses about deservingness and $redistSelf$, and we generate a measure of role-of-government beliefs using responses on private enterprise and $jobsForAll$. This yields two further outcome variables beyond our panel-study variables.\note{The precise survey questions are shown in Section~\ref{apdx:Exp:QuestionWordings}. We use the first PC in each case. Loadings are shown in section~\ref{apdx:Exp:PCALoadings}.}

Our primary theoretical interest is in establishing whether it was \emph{possible} for ideological rhetoric relating to COVID-19 to move people's attitudes. As such, the core hypothesis that we test is that, for each of our five dependent variables, the effect of at least one of our four treatment conditions is greater than zero. That is, for each DV:
\begin{equation} %\setlength\abovedisplayskip{0pt}
	\delta^{CRIP} > 0 \lor \delta^{URIP} > 0 \lor \delta^{SCP} > 0 \lor \delta^{SIP} > 0 ~,
	\label{hyp:Ftest}
\end{equation}
\noindent where the $\delta$s are treatment effects for each of the experimental primes.

Our secondary set of hypotheses relates to our expectation that the personal COVID-19 experience (PCE) treatment will moderate our treatment effects. As moderation effects are far more demanding in terms of sample size required for a given statistical power, we opted to collapse our four ideological-link treatments into a single treatment variable for this part of the analysis, the effect of which we denote as $\delta^{\mu}$. 
Hence we hypothesise that: 
\begin{equation}
\delta^{\mu}_{PCE} > \delta^{\mu}_{NPCE} \land \delta^{\mu}_{PCE} > 0,
\label{hyp:DeltaInteraction}
\end{equation}
\noindent where $\delta^{\mu}_{PCE}$ is the average treatment effect on left-right ideological beliefs when $PCE = 1$, and $\delta^{\mu}_{NPCE}$ is the average treatment effect when $PCE = 0$. 

Statistical estimation proceeds as follows. Let $\Omega \equiv \{CRIP, URIP, SIP, SCP\}$. Thus to test our main hypothesis set out in \cref{hyp:Ftest}, we estimate:
\begin{align}
	DV^{Post}_i	=	& \beta_0 + \beta_1 \cdot DV^{Pre}_i +  \sum_{\tau \in \Omega} \delta^{\tau} \tau_i + \epsilon_i \label{eq:BasePrePost} ~.
\end{align}
\noindent Our test then consists of an $F$-test for the joint significance of the $\delta$ coefficients. Because we conduct five such $F$-tests -- one for each dependent variable -- we adjust the $F$-test $p$-values for multiple comparisons using the \citet{BenjaminiHochberg1995} procedure.

To test the PCE-moderation hypothesis in \cref{hyp:DeltaInteraction}, let $T_i$ be a dummy equal to 1 if individual $i$ is in any of the ideological-link treatments. Then, we estimate:\note{See section~\ref{apdx:Exp:Results} for details.} 
\begin{equation}\label{eq:BasePrePostInteraction}
DV^{Post}_i = \beta_0 + \beta_1 \cdot DV^{Pre}_i + \delta^{\mu} T_i  + \delta^{PCE} PCE_i + \gamma (T_i \cdot PCE_i) + \epsilon_i  ~.
\end{equation}

The full results of our main hypothesis tests (unmoderated by PCE) are shown in Table~\ref{tab:experimentTreatmentEffects}, with the results moderated by PCE shown in Table~\ref{tab:experimentTreatmentInteractionEffects}. For our purposes, it is sufficient to note that across all treatment primes and all dependent variables, there is no evidence that our primes caused respondents to change their opinions on redistribution or the role of government. Figure~\ref{fig:experiment_results} shows the results without adjustments for multiple testing, but even here there is little evidence that the primes moved opinions. Furthermore, there is no sign that personal experience moderates the effect.

In section~\ref{apdx:Exp:TripleInteraction} we go further and evaluate whether our ideological-linkage treatments had differential effects for respondents who report having been personally affected by the COVID pandemic, either in terms of labor-market loss or receipt of government support. We find that even voters who were personally affected by the crisis, and were primed to think about their experiences, were largely unpersuaded by the ideological arguments we provided. In sum, within the limits of a survey experiment, we uncover no evidence that, had politicians used different discourse during the pandemic, opinions on redistribution and the role of government would have shifted.


\section*{Conclusion}
\label{sec:Conclusion}

We have found very little evidence that, in the UK, the pandemic or exposure to pandemic-related shocks shifted ideological beliefs to the left. We find minimal indication of changes in ideological beliefs even when looking in the places where such change should have been most likely to emerge: among those who personally experienced employment loss or receipt of benefits; among recipients of the most \emph{visible} forms of government support; or among relevant partisan subgroups. And we observe the same attitudinal stability both in the first few months of the pandemic and more than a year in. 

In examining possible explanations of the null effects, we rule out the possibility that the pandemic failed to generate ideological-belief change because it failed to change risk perceptions. We report strong evidence that those who lost employment or received government support substantially raised their estimates of the probability that they would experience unemployment or poverty over the next year. Results from our survey experiment undermine an explanation grounded in missed framing opportunities. More prominent elite efforts to connect pandemic experiences to broader ideological principles would have been unlikely to generate a leftward shift in beliefs. At the same time, the analysis leaves open the possibility that attitudinal stability in the midst of the pandemic is grounded in the stability of beliefs about the moral deservingness of those who receive benefits -- a set of mediating attitudes that we find changed little, even among those who themselves received benefits or experienced employment loss during the crisis.

We have argued that, from a theoretical perspective, the UK case featured conditions particularly conducive to attitude change in response to the COVID-19 crisis. In particular, the unprecedented expansion of the state's economic role, amid strong elite consensus on the necessity of that expansion, exposed the vast bulk of Britons to a much-altered and remarkably one-sided informational environment. The stability of ideological beliefs even in these circumstances suggests the pandemic is unlikely to have moved fundamental political orientations anywhere -- and certainly not in settings where the policy measures were more familiar or the level of elite disagreement higher. Indeed, prior studies examining opinion change elsewhere during the pandemic have generally reported little evidence of attitudinal shifts \citep{ReeskensEtAl2021,AresEtAl2021,JuradoKuo2022}.

The non-effects of the COVID-19 pandemic on public attitudes toward the role of the state suggest a possible irony in the way in which citizens process information about crises: the very features of crises that focus attention may also make them hard to exploit for the purposes of policy change. There can be little doubt that the pandemic drew attention to the health and economic problems that it generated, and there appears to have been broad support for the immediate governmental response. Yet it could be precisely \emph{because} of the unique severity of the coronavirus outbreak that the public declined to draw \emph{broader} ideological conclusions -- even when, as in our experiment, encouraged to do so. In other words, extraordinary events may, owing to the very extremity of their features, be seen by citizens as irrelevant to ``normal'' politics.

The pandemic's non-effect on attitudes toward the role of the state might also prompt us to reflect on the historical relationship between crises -- such as the Great Depression and World War II -- and mass ideology. As we have noted, historical accounts \autocite[e.g.,][]{Higgs2013, Titmuss2018} have depicted these events as reshaping mass ideological beliefs in advanced democracies, helping forge a popular consensus in favor of greater state intervention in the postwar economy \parencite{ObingerPetersen2017, WalterEmmenegger2022}.  The public opinion data needed to test these historical claims directly do not exist, and we do not wish to push the parallel between the pandemic and landmark 20th century events too far. Yet the results presented here lend support to an alternative, more historical-institutionalist account of the relationship between crisis and the rise of the postwar consensus. Historical institutionalists often understand major crises as ``critical junctures'' at which ``the range of plausible choices open to powerful political actors expands substantially and the consequences of their decisions for the outcome of interest are potentially much more momentous'' \autocite[343]{CapocciaKelemen2007}. On this alternative view, it would be elites who learned first, during periods of major upheaval, about the need for new forms of state intervention and who exploited the leeway afforded by crisis to enact policy transformations. While these elite-driven expansions of the state were typically welcomed by citizens, enduring change in public understandings of the state's role in society may have arisen gradually as a feedback effect of the programmatic (e.g., contributory) logic and lived enjoyment of new social entitlements \autocite{Pierson1993}. In other words, new mass ideological beliefs may have emerged not from direct experience of calamity and state support, but indirectly and over time through longer-term experience of policies established during crises. On this view, if the COVID-19 pandemic is to yield durable change in ideological beliefs, it will more likely emerge as a downstream \emph{effect} of elite investments in long-run policy change than as a driver of those decisions.

\section*{Acknowledgements}

We are grateful for previous feedback from participants at the CPEB seminar at the UCL Department of Political Science and the Max Planck Online Workshop in Comparative Political Economy. We also thank the British Election Study team for their generous assistance in fielding our surveys. 

%\clearpage
\printbibliography


\section*{Authors}

Jack Blumenau is an Associate Professor at University College London, WC1H 9QU, UK.

Timothy Hicks is an Associate Professor at University College London, WC1H 9QU, UK.

Alan M. Jacobs is a Professor at the University of British Columbia, Vancouver, BC Canada V6T 1Z1, Canada.

J. Scott Matthews is a Professor at Memorial University of Newfoundland, St. John’s, NL, A1C 3V3, Canada.

Tom O'Grady is an Associate Professor at University College London, WC1H 9QU, UK.


\end{document}